Video coding schemes, such as H.264/MPEG-4 AVC and the upcoming HEVC (High-Efficiency Video Coding) perform encoding/decoding of image/video content using inter-picture (or simply “inter”) prediction from previously encoded/decoded reference pictures to exploit information redundancy across consecutive pictures in time.
In a coded video bitstream, a reference picture used for the inter prediction process of a prediction unit (such as an M×N block of samples) is identified by or referred to using a reference index. A reference index is an index into an ordered list (known as a reference picture list) comprising one or more reference pictures. Each reference index is uniquely associated with a reference picture in the reference picture list. That is, the reference index is a value that is used to distinguish multiple reference pictures from each other.
The above-mentioned coding schemes support temporal prediction of motion vectors (i.e., motion vector prediction or MVP), whereby motion vectors of a target block of samples are predicted from the motion vectors of one or more previously coded blocks of samples in a collocated reference picture. Temporal motion vector prediction further reduces the bitrate associated with motion vectors by exploiting information redundancy among neighbouring motion vectors temporally. The collocated reference picture is selected among available reference pictures using a predetermined scheme, for example, selecting the first reference picture in a predetermined reference picture list (such as reference picture list 0) as the collocated reference picture.
In applications requiring video transmission across lossy environment, temporal motion vector prediction is susceptible to erroneous prediction of motion vector when the collocated reference picture is lost or contains errors. In the HEVC standard under development, a technique was disclosed for disabling temporal motion vector prediction of certain sub-picture unit (e.g., a slice). JCTVC-G398, “High-level Syntax: Marking process for non-TMVP pictures”, Joint Collaborative Team on Video Coding (JCT-VC) of ITU-T SG 16 WP3 and ISO/IEC JTC1/SC29/WG11 7th meeting, Geneva, CH, November 2011. In this technique, it is necessary to introduce a marking flag in a Picture Parameter Set (PPS) used for marking a picture in the Decoder Picture Buffer (DPB) as “unused for temporal motion vector prediction”. This marking process is performed by a decoder when a sub-picture unit refers to a PPS having a marking flag equal to “TRUE”.